genotypeGVCFs call confidence and emit thresholds
We have ~10,000 WES samples and generated gVCFs for each using HC. To generate a multi-sample consensus VCF, we performed joint genotyping using genotypeGVCFs. Subsequently we performed VQSR analysis to assign quality scores to variants. If VQSR fails variants, some of which have good quality scores in a few individual samples, can we rescue these by lowering emit or call confidence thresholds at the joint genotyping phase. If we can, how low can we go for these cutoffs?
[Just FYI, we wanted to test if we could rescue these variants by lowering emit and call confidence thresholds to 0 (which we know is a bad idea, allowing poor quality calls to creep in) . The number of variants more than doubled and variants that were initially not even detected received a VQSR "PASS", making the results highly unreliable.]